Prompt A Structured Q&A Response Using LLM Gateway

1
import requests
2
import time
3
import xml.etree.ElementTree as ET
4
5
API_KEY = "YOUR_API_KEY"
6
audio_url = "https://storage.googleapis.com/aai-web-samples/meeting.mp4"
7
8
# -------------------------------
9
# Step 1: Transcribe the audio
10
# -------------------------------
11
transcript_request = requests.post(
12
    "https://api.assemblyai.com/v2/transcript",
13
    headers={"authorization": API_KEY, "content-type": "application/json"},
14
    json={"audio_url": audio_url, "speech_models": ["universal-3-pro"]},
15
)
16
17
transcript_id = transcript_request.json()["id"]
18
19
# Poll for completion
20
while True:
21
    polling_response = requests.get(
22
        f"https://api.assemblyai.com/v2/transcript/{transcript_id}",
23
        headers={"authorization": API_KEY},
24
    )
25
    status = polling_response.json()["status"]
26
27
    if status == "completed":
28
        transcript_text = polling_response.json()["text"]
29
        break
30
    elif status == "error":
31
        raise RuntimeError(f"Transcription failed: {polling_response.json()['error']}")
32
    else:
33
        print(f"Transcription status: {status}")
34
        time.sleep(3)
35
36
# -------------------------------
37
# Step 2: Build question helper functions
38
# -------------------------------
39
def construct_question(question):
40
    question_str = f"Question: {question['question']}"
41
42
    if question.get("context"):
43
        question_str += f"\nContext: {question['context']}"
44
45
    # Default answer_format
46
    if not question.get("answer_format"):
47
        question["answer_format"] = "short sentence"
48
49
    question_str += f"\nAnswer Format: {question['answer_format']}"
50
51
    if question.get("answer_options"):
52
        options_str = ", ".join(question["answer_options"])
53
        question_str += f"\nOptions: {options_str}"
54
55
    return question_str + "\n"
56
57
def escape_xml_characters(xml_string):
58
    return xml_string.replace("&", "&")
59
60
# -------------------------------
61
# Step 3: Define questions
62
# -------------------------------
63
questions = [
64
    {
65
        "question": "What are the top level KPIs for engineering?",
66
        "context": "KPI stands for key performance indicator",
67
        "answer_format": "short sentence",
68
    },
69
    {
70
        "question": "How many days has it been since the data team has gotten updated metrics?",
71
        "answer_options": ["1", "2", "3", "4", "5", "6", "7", "more than 7"],
72
    },
73
    {"question": "What are the future plans for the project?"},
74
]
75
76
question_str = "\n".join(construct_question(q) for q in questions)
77
78
# -------------------------------
79
# Step 4: Build the LLM prompt
80
# -------------------------------
81
prompt = f"""You are an expert at giving accurate answers to questions about texts.
82
No preamble.
83
Given the series of questions, answer the questions.
84
Each question may follow up with answer format, answer options, and context for each question.
85
It is critical that you follow the answer format and answer options for each question.
86
When context is provided with a question, refer to it when answering the question.
87
You are useful, true and concise, and write in perfect English.
88
Only the question is allowed between the <question> tag. Do not include the answer format, options, or question context in your response.
89
Only text is allowed between the <question> and <answer> tags.
90
XML tags are not allowed between the <question> and <answer> tags.
91
End your response with a closing </responses> tag.
92
For each question-answer pair, format your response according to the template provided below:
93
94
Template for response:
95
<responses>
96
  <response>
97
    <question>The question</question>
98
    <answer>Your answer</answer>
99
  </response>
100
  <response>
101
    ...
102
  </response>
103
  ...
104
</responses>
105
106
These are the questions:
107
{question_str}
108
109
Transcript:
110
{transcript_text}
111
"""
112
113
# -------------------------------
114
# Step 5: Query LLM Gateway
115
# -------------------------------
116
headers = {"authorization": API_KEY}
117
118
response = requests.post(
119
    "https://llm-gateway.assemblyai.com/v1/chat/completions",
120
    headers=headers,
121
    json={
122
        "model": "claude-sonnet-4-5-20250929",
123
        "messages": [{"role": "user", "content": prompt}],
124
        "max_tokens": 2000,
125
    },
126
)
127
128
response_json = response.json()
129
llm_output = response_json["choices"][0]["message"]["content"]
130
131
# -------------------------------
132
# Step 6: Parse and print XML response
133
# -------------------------------
134
clean_response = escape_xml_characters(llm_output).strip()
135
136
try:
137
    root = ET.fromstring(clean_response)
138
    for resp in root.findall("response"):
139
        question = resp.find("question").text
140
        answer = resp.find("answer").text
141
        print(f"Question: {question}")
142
        print(f"Answer: {answer}\n")
143
except ET.ParseError as e:
144
    print("Could not parse XML response.")
145
    print("Raw model output:\n", llm_output)

1
import requests
2
import time
3
import xml.etree.ElementTree as ET
4
5
API_KEY = "YOUR_API_KEY"

1
audio_url = "https://storage.googleapis.com/aai-web-samples/meeting.mp4"
2
3
# -------------------------------
4
# Step 1: Transcribe the audio
5
# -------------------------------
6
transcript_request = requests.post(
7
    "https://api.assemblyai.com/v2/transcript",
8
    headers={"authorization": API_KEY, "content-type": "application/json"},
9
    json={"audio_url": audio_url, "speech_models": ["universal-3-pro"]},
10
)
11
12
transcript_id = transcript_request.json()["id"]
13
14
# Poll for completion
15
while True:
16
    polling_response = requests.get(
17
        f"https://api.assemblyai.com/v2/transcript/{transcript_id}",
18
        headers={"authorization": API_KEY},
19
    )
20
    status = polling_response.json()["status"]
21
22
    if status == "completed":
23
        transcript_text = polling_response.json()["text"]
24
        break
25
    elif status == "error":
26
        raise RuntimeError(f"Transcription failed: {polling_response.json()['error']}")
27
    else:
28
        print(f"Transcription status: {status}")
29
        time.sleep(3)

1
# -------------------------------
2
# Step 2: Build question helper functions
3
# -------------------------------
4
def construct_question(question):
5
    question_str = f"Question: {question['question']}"
6
7
    if question.get("context"):
8
        question_str += f"\nContext: {question['context']}"
9
10
    # Default answer_format
11
    if not question.get("answer_format"):
12
        question["answer_format"] = "short sentence"
13
14
    question_str += f"\nAnswer Format: {question['answer_format']}"
15
16
    if question.get("answer_options"):
17
        options_str = ", ".join(question["answer_options"])
18
        question_str += f"\nOptions: {options_str}"
19
20
    return question_str + "\n"
21
22
def escape_xml_characters(xml_string):
23
    return xml_string.replace("&", "&")

1
# -------------------------------
2
# Step 3: Define questions
3
# -------------------------------
4
questions = [
5
    {
6
        "question": "What are the top level KPIs for engineering?",
7
        "context": "KPI stands for key performance indicator",
8
        "answer_format": "short sentence",
9
    },
10
    {
11
        "question": "How many days has it been since the data team has gotten updated metrics?",
12
        "answer_options": ["1", "2", "3", "4", "5", "6", "7", "more than 7"],
13
    },
14
    {"question": "What are the future plans for the project?"},
15
]

1
question_str = '\n'.join(construct_question(q) for q in questions)

1
# -------------------------------
2
# Step 4: Build the LLM prompt
3
# -------------------------------
4
prompt = f"""You are an expert at giving accurate answers to questions about texts.
5
No preamble.
6
Given the series of questions, answer the questions.
7
Each question may follow up with answer format, answer options, and context for each question.
8
It is critical that you follow the answer format and answer options for each question.
9
When context is provided with a question, refer to it when answering the question.
10
You are useful, true and concise, and write in perfect English.
11
Only the question is allowed between the <question> tag. Do not include the answer format, options, or question context in your response.
12
Only text is allowed between the <question> and <answer> tags.
13
XML tags are not allowed between the <question> and <answer> tags.
14
End your response with a closing </responses> tag.
15
For each question-answer pair, format your response according to the template provided below:
16
17
Template for response:
18
<responses>
19
  <response>
20
    <question>The question</question>
21
    <answer>Your answer</answer>
22
  </response>
23
  <response>
24
    ...
25
  </response>
26
  ...
27
</responses>
28
29
These are the questions:
30
{question_str}
31
32
Transcript:
33
{transcript_text}
34
"""
35
36
# -------------------------------
37
# Step 5: Query LLM Gateway
38
# -------------------------------
39
headers = {"authorization": API_KEY}
40
41
response = requests.post(
42
    "https://llm-gateway.assemblyai.com/v1/chat/completions",
43
    headers=headers,
44
    json={
45
        "model": "claude-sonnet-4-5-20250929",
46
        "messages": [{"role": "user", "content": prompt}],
47
        "max_tokens": 2000,
48
    },
49
)
50
51
response_json = response.json()
52
llm_output = response_json["choices"][0]["message"]["content"]
53
54
# -------------------------------
55
# Step 6: Parse and print XML response
56
# -------------------------------
57
clean_response = escape_xml_characters(llm_output).strip()
58
59
try:
60
    root = ET.fromstring(clean_response)
61
    for resp in root.findall("response"):
62
        question = resp.find("question").text
63
        answer = resp.find("answer").text
64
        print(f"Question: {question}")
65
        print(f"Answer: {answer}\n")
66
except ET.ParseError as e:
67
    print("Could not parse XML response.")
68
    print("Raw model output:\n", llm_output)